ext4: Use atomic_t's in struct flex_groups

Reduce pressure on the sb_bgl_lock family of locks by using atomic_t's
to track the number of free blocks and inodes in each flex_group.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index b37b128..53c72ad 100644 (file)
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -470,9 +470,8 @@ void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
 
        if (sbi->s_log_groups_per_flex) {
                ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
-               spin_lock(sb_bgl_lock(sbi, flex_group));
-               sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
-               spin_unlock(sb_bgl_lock(sbi, flex_group));
+               atomic_add(blocks_freed,
+                          &sbi->s_flex_groups[flex_group].free_blocks);
        }
        /*
         * request to reload the buddy with the

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index e5c273f..e52b48f 100644 (file)
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -170,8 +170,8 @@ struct ext4_group_desc
  */
 
 struct flex_groups {
-       __u32 free_inodes;
-       __u32 free_blocks;
+       atomic_t free_inodes;
+       atomic_t free_blocks;
 };
 
 #define EXT4_BG_INODE_UNINIT   0x0001 /* Inode table/bitmap not in use */

diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 617f5a2..5f39392 100644 (file)
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -189,7 +189,6 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
        struct ext4_super_block *es;
        struct ext4_sb_info *sbi;
        int fatal = 0, err, count, cleared;
-       ext4_group_t flex_group;
 
        if (atomic_read(&inode->i_count) > 1) {
                printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
@@ -277,10 +276,10 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
                                percpu_counter_dec(&sbi->s_dirs_counter);
 
                        if (sbi->s_log_groups_per_flex) {
-                               flex_group = ext4_flex_group(sbi, block_group);
-                               spin_lock(sb_bgl_lock(sbi, flex_group));
-                               sbi->s_flex_groups[flex_group].free_inodes++;
-                               spin_unlock(sb_bgl_lock(sbi, flex_group));
+                               ext4_group_t f;
+
+                               f = ext4_flex_group(sbi, block_group);
+                               atomic_inc(&sbi->s_flex_groups[f].free_inodes);
                        }
                }
                BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
@@ -360,9 +359,9 @@ static int find_group_flex(struct super_block *sb, struct inode *parent,
                sbi->s_log_groups_per_flex;
 
 find_close_to_parent:
-       flexbg_free_blocks = flex_group[best_flex].free_blocks;
+       flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
        flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
-       if (flex_group[best_flex].free_inodes &&
+       if (atomic_read(&flex_group[best_flex].free_inodes) &&
            flex_freeb_ratio > free_block_ratio)
                goto found_flexbg;
 
@@ -375,24 +374,24 @@ find_close_to_parent:
                if (i == parent_fbg_group || i == parent_fbg_group - 1)
                        continue;
 
-               flexbg_free_blocks = flex_group[i].free_blocks;
+               flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
                flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 
                if (flex_freeb_ratio > free_block_ratio &&
-                   flex_group[i].free_inodes) {
+                   (atomic_read(&flex_group[i].free_inodes))) {
                        best_flex = i;
                        goto found_flexbg;
                }
 
-               if (flex_group[best_flex].free_inodes == 0 ||
-                   (flex_group[i].free_blocks >
-                    flex_group[best_flex].free_blocks &&
-                    flex_group[i].free_inodes))
+               if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
+                   ((atomic_read(&flex_group[i].free_blocks) >
+                     atomic_read(&flex_group[best_flex].free_blocks)) &&
+                    atomic_read(&flex_group[i].free_inodes)))
                        best_flex = i;
        }
 
-       if (!flex_group[best_flex].free_inodes ||
-           !flex_group[best_flex].free_blocks)
+       if (!atomic_read(&flex_group[best_flex].free_inodes) ||
+           !atomic_read(&flex_group[best_flex].free_blocks))
                return -1;
 
 found_flexbg:
@@ -960,9 +959,7 @@ got:
 
        if (sbi->s_log_groups_per_flex) {
                flex_group = ext4_flex_group(sbi, group);
-               spin_lock(sb_bgl_lock(sbi, flex_group));
-               sbi->s_flex_groups[flex_group].free_inodes--;
-               spin_unlock(sb_bgl_lock(sbi, flex_group));
+               atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
        }
 
        inode->i_uid = current_fsuid();

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index c4c4309..e72c72a 100644 (file)
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -3044,9 +3044,8 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
        if (sbi->s_log_groups_per_flex) {
                ext4_group_t flex_group = ext4_flex_group(sbi,
                                                          ac->ac_b_ex.fe_group);
-               spin_lock(sb_bgl_lock(sbi, flex_group));
-               sbi->s_flex_groups[flex_group].free_blocks -= ac->ac_b_ex.fe_len;
-               spin_unlock(sb_bgl_lock(sbi, flex_group));
+               atomic_sub(ac->ac_b_ex.fe_len,
+                          &sbi->s_flex_groups[flex_group].free_blocks);
        }
 
        err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
@@ -4884,9 +4883,7 @@ do_more:
 
        if (sbi->s_log_groups_per_flex) {
                ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
-               spin_lock(sb_bgl_lock(sbi, flex_group));
-               sbi->s_flex_groups[flex_group].free_blocks += count;
-               spin_unlock(sb_bgl_lock(sbi, flex_group));
+               atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks);
        }
 
        ext4_mb_release_desc(&e4b);

diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index c06886a..546c7dd 100644 (file)
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -938,10 +938,10 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
                ext4_group_t flex_group;
                flex_group = ext4_flex_group(sbi, input->group);
-               sbi->s_flex_groups[flex_group].free_blocks +=
-                       input->free_blocks_count;
-               sbi->s_flex_groups[flex_group].free_inodes +=
-                       EXT4_INODES_PER_GROUP(sb);
+               atomic_add(input->free_blocks_count,
+                          &sbi->s_flex_groups[flex_group].free_blocks);
+               atomic_add(EXT4_INODES_PER_GROUP(sb),
+                          &sbi->s_flex_groups[flex_group].free_inodes);
        }
 
        ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);

diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 1ec554c..6b5d5c6 100644 (file)
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1630,10 +1630,10 @@ static int ext4_fill_flex_info(struct super_block *sb)
                gdp = ext4_get_group_desc(sb, i, &bh);
 
                flex_group = ext4_flex_group(sbi, i);
-               sbi->s_flex_groups[flex_group].free_inodes +=
-                       ext4_free_inodes_count(sb, gdp);
-               sbi->s_flex_groups[flex_group].free_blocks +=
-                       ext4_free_blks_count(sb, gdp);
+               atomic_set(&sbi->s_flex_groups[flex_group].free_inodes,
+                          ext4_free_inodes_count(sb, gdp));
+               atomic_set(&sbi->s_flex_groups[flex_group].free_blocks,
+                          ext4_free_blks_count(sb, gdp));
        }
 
        return 1;